沐曦 Metax GPU 服务器上编译失败

参考 https://github.com/PaddlePaddle/community/blob/master/pfcc/paddle-hardware/%E6%B2%90%E6%9B%A6-FastDeploy-%E7%BC%96%E8%AF%91%E9%AA%8C%E8%AF%81-%E6%89%93%E5%8D%A1%E4%BB%BB%E5%8A%A1.md

https://gitee.com/paddlepaddle/FastDeploy/blob/release/2.4/docs/zh/get_started/installation/metax_gpu.md

在沐曦的服务器上编译失败（使用 release/2.4 和 develop 都是同样的错误），提示很多错误，这里只摘取部分：

/data/FastDeploy/custom_ops/gpu_ops/speculate_decoding/speculate_clear_accept_nums.cu compile failed, command '/usr/local/cuda/bin/nvcc' failed: No such file or directory
In file included from /data/FastDeploy/custom_ops/gpu_ops/remote_cache_kv_ipc.cc:15:
/data/FastDeploy/custom_ops/gpu_ops/remote_cache_kv_ipc.h:28:10: fatal error: driver_types.h: No such file or directory
   28 | #include "driver_types.h"
      |          ^~~~~~~~~~~~~~~~ [2026-03-28 10:02:04,113] [    INFO] spawn.py:38 - gcc -pthread -B /opt/conda/compiler_compat -Wno-unused-result -Wsign-compare -DNDEBUG -fwrapv -O2 -Wall -fPIC -O2 -isystem /opt/conda/include -fPIC -O2 -isystem /opt/conda/include -fPIC -I/opt/maca/include -I/opt/maca/include/mcr -I/opt/maca/include/common -I/opt/maca/include/mcfft -I/opt/maca/include/mcrand -I/opt/maca/include/mcsparse -I/opt/maca/include/mcblas -I/opt/maca/include/mcsolver -I/opt/conda/lib/python3.10/site-packages/paddle/include -I/opt/conda/lib/python3.10/site-packages/paddle/include/third_party -I/opt/conda/lib/python3.10/site-packages/paddle/include/paddle/phi/api/include/compat -I/opt/conda/lib/python3.10/site-packages/paddle/include/paddle/phi/api/include/compat/torch/csrc/api/include -I/opt/conda/include/python3.10 -I/opt/conda/include/python3.10 -c /data/FastDeploy/custom_ops/gpu_ops/speculate_decoding/speculate_update_input_ids_cpu.cc -o /data/FastDeploy/custom_ops/build/fastdeploy_ops/lib.linux-x86_64-cpython-310/build/fastdeploy_ops/temp.linux-x86_64-cpython-310/gpu_ops/speculate_decoding/speculate_update_input_ids_cpu.o -O3 -w -DPADDLE_WITH_CUSTOM_KERNEL -DPADDLE_EXTENSION_NAME=fastdeploy_ops -D_GLIBCXX_USE_CXX11_ABI=1 -DPADDLE_WITH_CUDA -std=c++17
/data/FastDeploy/custom_ops/gpu_ops/speculate_decoding/speculate_set_stop_value_multi_seqs.cu compile failed, command '/usr/local/cuda/bin/nvcc' failed: No such file or directory
In file included from /opt/conda/lib/python3.10/site-packages/paddle/include/paddle/phi/api/include/api.h:5,
                 from /opt/conda/lib/python3.10/site-packages/paddle/include/paddle/phi/api/all.h:24,
                 from /opt/conda/lib/python3.10/site-packages/paddle/include/paddle/extension.h:29,
                 from /data/FastDeploy/custom_ops/gpu_ops/get_img_boundaries.cc:15:
/opt/conda/lib/python3.10/site-packages/paddle/include/paddle/phi/api/include/tensor.h:23:10: fatal error: cuda_runtime.h: No such file or directory
   23 | #include <cuda_runtime.h>
      |          ^~~~~~~~~~~~~~~~